What's new:
- fix draw detection
- fix history overflow
- fix mate score for probcut
- use fifty move guard on tt in qsearch
- update piece stack for null move pruning
- improve tt replace strategy
- fix ply coherence in singular search
- fix undefined behavior in history
- fix undo bug
- apply incheck futility guard
- use tt for skipmove positions
- order moves during probcut
- detect bmi2 in basic makefile
- revert standard network back to c049c117
- support vnni256
- optimize quiets pickup
- speed-up castling
- store tt wdl result as a bonus
- fix tb probing with wrong castling rights
- reduce tt captures
- minimize see calls
- implement finny table
- implement counter move table
- track best move stability
- use tt for null moves
- release new standard network
- implement evalfile parameter
- fix tb tt bug
- restore hce build for tcec
- probe dtz
- fix tt castling bit for frc
- cache hash entry
- speed-up inferencing code
- implement avx512 support
- improve voting system
- fix gcc14 build
- implement frc mode
- improve speed
- support longer games
- optimize sort
- penalize bad captures
Regression run against Igel 3.6.0 at Long Time Control (60+0.6):
Score of Igel 3.7.0 64 BMI2 AVX512 VNNI512 vs Igel 3.6.0 64 BMI2 AVX2: 6161 - 1832 - 7449 [0.640] 15442
Elo difference: 100.1 +/- 3.9, LOS: 100.0 %, DrawRatio: 48.2 %
Regression run against Igel 3.6.0 at Short Time Control (10+0.1):
Score of Igel 3.7.0 64 BMI2 AVX512 VNNI512 vs Igel 3.6.0 64 BMI2 AVX2: 7122 - 2188 - 7167 [0.650] 16477
Elo difference: 107.3 +/- 4.0, LOS: 100.0 %, DrawRatio: 43.5 %